How an agent works: the simple loop

You don't need any technical background to understand how an AI agent works, because it comes down to a simple repeating loop: perceive → reason → act.

• Perceive — the agent reads the situation: your request, plus whatever it's looking at (a web page, a file, the result of its last step).
• Reason — it decides what to do next: which step to take, or which tool to use.
• Act — it takes that step (searches, writes, clicks, runs something).

Then it loops back: it perceives the result of its action, reasons about what to do next, and acts again — repeating until the goal is done, or until it gets stuck and needs your help.

A concrete example — a "research agent." Ask one to research a topic, and it loops: plan what to look for → search → read a source → decide it needs more → search again → read → synthesize → write a summary with citations. It might repeat that loop across dozens of sources, all on its own, before handing you a finished report. That's the loop in action.

Why the loop matters for understanding agents:

• It's why agents can do multi-step work a chatbot can't — they keep going, using each result to decide the next move.
• It's also why they can go wrong — a mistake early in the loop gets carried forward and can compound over many steps (the next lesson explains why this makes long tasks unreliable).
• It's why tools matter — an agent is only as capable as the tools it can use (search, code, apps). More tools = more it can do, but also more it can affect (which is where risk comes in — Module 2).

The mindset: an AI agent runs a simple loop — perceive the situation, reason about the next step, act, then repeat — using tools and reacting to results until the job is done. That loop is what lets agents tackle multi-step tasks a chatbot can't, and understanding it demystifies the whole "agentic AI" idea. It also hints at both the promise (they keep working toward a goal) and the pitfalls (errors compound, and they can affect real things) that the rest of this course covers.